Towards better understanding of frequent itemset relationships through tree-like data structures

摘要

A common goal of descriptive data mining techniques is presenting new information in concise, easily interpretable and understandable ways. In this paper we propose a technique for modeling relationships between frequent itemsets through visually descriptive tree-like data structures. We define and discuss algorithms for forming these structures as well as suggest new measures for evaluating their informative value. We also present our visualization tool which implements proposed concepts and solutions. Finally, we apply our research on two different dataset types and discuss the results. The first dataset proves the applicability of our visualization technique for common market basket analysis. The second dataset is an example of a "dense" dataset, a troublesome type for frequent itemset mining since it commonly produces a significantly large number of frequent itemsets. We demonstrate a modified variant of our technique which allows efficient visual representation of such datasets as well.